Biology and Compositionality: Empirical Considerations for Emergent-Communication Protocols

11/26/2019 ∙ by Travis LaCroix, et al. ∙ University of California, Irvine 0

Significant advances have been made in artificial systems by using biological systems as a guide. However, there is often little interaction between computational models for emergent communication and biological models of the emergence of language. Many researchers in language origins and emergent communication take compositionality as their primary target for explaining how simple communication systems can become more like natural language. However, there is reason to think that compositionality is the wrong target on the biological side, and so too the wrong target on the machine-learning side. As such, the purpose of this paper is to explore this claim. This has theoretical implications for language origins research more generally, but the focus here will be the implications for research on emergent communication in computer science and machine learning—specifically regarding the types of programmes that might be expected to work and those which will not. I further suggest an alternative approach for future research which focuses on reflexivity, rather than compositionality, as a target for explaining how simple communication systems may become more like natural language. I end by providing some reference to the language origins literature that may be of some use to researchers in machine learning.



There are no comments yet.


page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Communication is ubiquitous in nature, but linguistic communication—i.e., natural language—is unique to humans. There are inherent difficulties in determining how language may have evolved—i.e., out of simpler communicative precursors (proto-language). On the one hand, it is difficult to define what constitutes language—it is always in flux, it is infinitely flexible, and it is one of the most complex behaviours known (Christiansen and Kirby, 2003). On the other hand, there is a paucity of direct evidence available with which to confirm current theories—language does not fossilise, nor can we go back in time to observe the actual precursors of human-level linguistic capacities.

Language origins research may appear to be a purely epistemic project; however, this programme finds application in contemporary machine learning and artificial intelligence (ML/AI). Understanding the fundamental principles that are involved in the (biological and cultural) evolution of effective communication may lead to innovative communication methods for use by interacting AI agents and multi-robot systems

(Wagner et al., 2003). AI agents need a common language to coordinate successfully and to communicate with humans. Furthermore, given the close relationship between language and thought, a clear understanding of how languages arise and persist in a population may also lend some additional insight into human cognitive capacities, helping lead the way in creating an artificial general intelligence (AGI).

The similar goals of language origins research (from a philosophical, biological, bio-linguistic, and cognitive systems perspective) and emergent communication research (from a computational perspective) should be apparent. One of the questions asked in contemporary machine learning is

How can emergent communication become more like natural language?

This question finds parallel expression in language origins research:

How might something like natural language evolve out of a simple communication system?

Thus, progress in language origins may help direct research in emergent communication and vice-versa. However, there has been relatively little dialogue between these two approaches.

To explain how language evolved out of simpler precursors, it is necessary to understand the salient differences between language and simple communication, or signalling. Most researchers (in both programmes) working on these questions take compositionality (productivity, openness, hierarchical structure, generative capacity, complex syntax, etc.) as their primary target. Demonstrating how compositionality may have evolved is supposed to be (at least partially) sufficient for explaining how language evolved. Furthermore, narrowing one’s explanatory target from the formidable question of how language evolved to the question of how one component of language—e.g., complex syntax—evolved affords some degree of analytic tractability.

However, there is reason to think that compositionality is the wrong target, on the biological side (and so the wrong target on the ML side). As such, the purpose of this paper is to explore this claim. This has theoretical implications for language origins research more generally, but the focus here will be the implications for research on emergent communication in computer science and ML—specifically regarding the types of programmes that might be expected to work, and those which will not. My hope, then, is that this work will help to direct future research in fruitful new directions.

2 A Quick Note on Compositionality

Compositionality provides a prima facie plausible explanatory target for language origins and emergent communication research. In the former case, this is because compositional syntax is a salient differentiating feature of language which is absent in simple communication systems found in nature. The principle of compositionality is typically stated as follows: The meaning of a compound [complex] expression is a function of the meaning of its parts [constituents] and how they are combined [composed] (Kamp and Partee, 1995; Szabó, 2012). In ML/AI, this is usually cashed out as a notion of systematicity; namely, “the ability to entertain a given thought implies the ability to entertain thoughts with semantically related contents” (Fodor and Pylyshyn, 1988, 3).

Any attempt to give an evolutionary explanation of human-level linguistic capacities will minimally need to account for the following: (1) how compositionality might arise from non-compositional communication; (2) if compositionality itself is an evolutionary adaptation, why compositional structure should be selected for in the first place; (3) why compositionality should be rare in nature, though communication is universal. Several models111See Nowak and Krakauer (1999); Barrett (2006, 2007, 2009); Franke (2014, 2016); Steinert-Threlkeld (2014, 2016, 2018); Barrett et al. (2018). have been suggested in recent years which grapple with these questions using the signalling-game framework.222The signalling game was introduced in Lewis (1969) and extended to dynamic models in Skyrms (1996, 2010). In the ML literature, this is sometimes called a referential game; see Lazaridou et al. (2016); Das et al. (2017); Evtimova et al. (2017); Havrylov and Titov (2017); Kottur et al. (2017); Choi et al. (2018); Lazaridou et al. (2018). Signalling games show how straightforward communication conventions might arise naturally through processes of repeated interactions. In an evolutionary context, starting with initially random signals and actions, individuals in a population learn or evolve effective communication conventions under several different dynamics.

However, LaCroix (2019a) suggests that the evolutionary explanations for compositional signalling offered thus far fail to give a plausible account of how compositionality might arise. The reason for this failure is twofold. On the one hand, these models often (if implicitly) take compositionality qua linguistic compositionality as their target for an evolutionary explanation. This gives rise to significant complications insofar as linguistic compositionality is rife with conceptual difficulties. On the other hand, these models fail to take into account the role asymmetry of the sender and receiver in the signalling game and thus fail to capture how compositionality might be beneficial for communication. LaCroix (2019a) suggests that if compositionality is a good target of language origins (or emergent communication) research, then it is necessary to build a more straightforward notion of compositionality for complex signalling from the bottom-up. Here, I suggest that compositionality is not even the right target for these research programmes.

3 Language Origins

In the context of the salient distinctions between language and animal communication systems, the problem is that compositional signals are extremely rare, or nonexistent, in nature. There is scant empirical evidence for call sequences that are compositionally meaningful. Most current data that suggests compositional signalling in nature comes from Zuberbühler’s study of several species of African forest monkeys (Cercopithecus) (Zuberbühler, 2001, 2002; Arnold and Zuberbühler, 2006a, b, 2008, 2013). However, no such combinatorial capacity or anything similar has been found in any other species of monkey or great ape, though these are well-studied (Fitch, 2010).

As a result, there is little evidence of a precursor to human-level compositional syntax. The most recent last common ancestor (LCA) between the great apes (including humans) and old-world monkeys (including Cercopithecus) existed approximately mya (million years ago), during the Paleogene period. Since no other related species utilises such compositional capacities, it stands to reason that this syntactic disposition was not present in the LCA of Homo and Cercopithecus—the most recent common ancestor of humans and chimpanzees, for example, lived around mya. Australopithecine species evolved after this break, with Homo likely evolving approximately mya.

Suppose that the LCA to great apes and old-world monkeys did have functionally proto-compositional syntax. This requires that proto-compositional syntax evolved up to mya. Then, we should expect that most old-world monkeys and great apes would show signs of using proto-compositional syntax. However, as far as we can tell, they do not—excepting the few mentioned Cercopithecus species. Assuming these dispositions evolved in the LCA, mya, and given that they do not appear in most decedents of the LCA, this implies that this disposition was lost—and lost more readily than it was held onto—in almost all other Catarrhine species.

Further still, it is controversial whether Neanderthals had language ( mya), let alone whether H. heidelbergensis had language ( mya). But, it is generally accepted that Australopithecine species ( mya) did not have language (Fitch, 2010). This implies that this proto-compositional disposition evolved into fully-fledged natural language on the order of mya.333Berwick and Chomsky (2016)

give a more conservative estimate, between

mya, corresponding to the first anatomically-modern humans and the last exodus from Africa. Meanwhile, these dispositions in few Catarrhine species remained utterly unchanged for more than million years, and these dispositions were lost in almost all great apes and old-world monkeys. This is the implausible (though technically not impossible) conclusion that follows from our assumption that a proto-compositional disposition existed in the LCA of great apes and old-world monkeys.

Thus, the scant empirical evidence for compositional precursors to human language comes paired with an improbable evolutionary history. The more plausible alternative is that there is no empirical evidence for any precursor to human language in nature. Note that an account which posits the sudden emergence of compositional syntax does not fall prey to these criticisms since it requires no proto-compositional precursor to compositional language. This is a virtue of the so-called saltationist perspective on language origins. However, this view is not without its own problems (LaCroix, 2019d). Therefore, if gradualism is the correct approach to language origins, then compositionality cannot be a correct target.

4 Consequences for Machine Learning

Significant advances have been made in artificial systems by using biological systems as a guide. For example, the development of reinforcement learning algorithms was heavily inspired by learning rules that were studied empirically in biology

(Bush and Mosteller, 1955; Rescorla and Wagner, 1972; Roth and Erev, 1995; Erev and Roth, 1998). A recent review of points of contact between ML and biology suggests several different areas of future research which can benefit from a bidirectional flow of information between these fields (Neftci and Averbeck, 2019); however, nowhere is mentioned the interface between biological and computational models of the emergence of language.

I suggested above that a gradualist approach to language origins in biology implies that compositionality is an incorrect target for this research programme. Further, computer scientists working on emergent communication also take compositionality as a primary target—a benchmark for success. However, recent work in ML highlights several problems for learning compositional linguistic structures. In particular, neural networks latch on to statistical regularities in existing datasets:

Bahdanau et al. (2018) highlight that, in a synthetic instruction-following task (Lake and Baroni, 2017), the agent does not learn a generalisation for composing words. Thus, when the agent is trained on the commands ‘jump’, ‘run twice’, and ‘walk twice’, it subsequently fails when asked to interpret ‘jump twice’. Thus, they fail to generalise. The above analysis suggests a possible reason why they do so fail: the target is compositional communication. Thus, if ML is to take cues from biological systems, it seems that this is not the correct direction for research in emergent communication.

5 Where to Go From Here: Reflexivity

Communication is a unique evolutionary process in the following sense: once a group of individuals has learned some set of simple communication conventions, those learned behaviours may be used to influence future communicative behaviours, giving rise to a feedback loop. When faced with a novel context, an individual can always learn a brand new disposition. However, the individual may learn to take advantage of previously evolved dispositions. Indeed, they may learn to take advantage of pre-evolved communicative dispositions to thereby influence the evolution of future communicative dispositions. This is a notion of reflexivity. Reflexivity, unlike compositionality, is consistent with a gradualist approach to language origins. Once actors exhibit such complexity, at a small scale, it may lead to a feedback loop between communication and cognition that, over time, gives rise to the complexity that we see in natural languages. Thus, this evolutionary story depends inherently on a notion of conceptual bootstrapping (Carey, 2004, 2009a, 2009b, 2011a, 2011b, 2014; Shea, 2009; Margolis and Laurence, 2008, 2011; Piantadosi et al., 2012; Beck, 2017).

To improve performance on generalisation, researchers in ML might add modularity and structure to their designs (Andreas et al., 2016; Gaunt et al., 2016). In the case of the Neural Module Network paradigm, a neural network is assembled from several modules, each of which is supposed to perform a particular subtask of the input processing. Bahdanau et al. (2018) note that although this modular approach is intuitively appealing, “widespread adoption has been hindered by the large amount of domain knowledge that is required to decide or predict how the modules should be created …and how they should be connected …based on a natural language utterance”. See also, Andreas et al. (2016); Johnson et al. (2016, 2017); Hu et al. (2017). Insofar as reflexivity is an apt target for the biological evolution of linguistic communication, it may too provide some insights for modelling emergent communication in an artificial system. See further discussion in LaCroix (2019c, 2020).

6 Further Resources

The gradualist view (though not necessarily via natural selection) is endorsed by, e.g., Givón (1979, 2002a, 2002b, 2009); Pinker and Bloom (1990); Newmeyer (1991, 1998, 2005); Jackendoff (1999, 2002); Carstairs-McCarthy (1999); Fitch (2004, 2010); Culicover and Jackendoff (2005); Számadó and Szathmary (2006); Progovac (2006, 2009a, 2009b, 2013, 2015, 2019); Tallerman (2007, 2013a, 2013b, 2014a, 2014b); Heine and Kuteva (2007); Hurford (2007, 2012); Tallerman and Gibson (2011); Yang (2013), among many others. The saltationist view is endorsed by Berwick (1998); Berwick et al. (2013a, b); Bickerton (1990, 1998); Lightfoot (1991); Hauser et al. (2002); Chomsky (2002, 2005, 2010); Piattelli-Palmarini and Uriagereka (2004, 2011); Moro (2008); Hornstein (2008); Piattelli-Palmarini (2010); Berwick and Chomsky (2011, 2016); Di Sciullo (2011, 2013); Bolhuis et al. (2014); Miyagawa et al. (2015); Miyagawa (2017), etc.

This is to say nothing of the signalling-game framework (Lewis, 1969; Skyrms, 1996, 2010)

in evolutionary game theory, which has seen a number of significant advances in a variety of philosophically interesting domains. These include, e.g., the difference between indicatives and imperatives

(Huttegger, 2007; Zollman, 2011); signalling in social dilemmas (Wagner, 2014); network formation (Pemantle and Skyrms, 2004; Barrett et al., 2017); deception (Zollman et al., 2012; Martínez, 2015; Skyrms and Barrett, 2018)

; meta-linguistic notions of truth and probability

(Barrett, 2016, 2017); syntactic structure and compositionality (Franke, 2016; Steinert-Threlkeld, 2016; Barrett et al., 2018; LaCroix, 2019e); vagueness (O’Connor, 2014); and epistemic representations, such as how the structure of one’s language evolves to maintain sensitivity to the structure of the world (Barrett and LaCroix, 2018). See LaCroix (2019b) for an overview.


  • Andreas et al. (2016) Andreas, Jacob, Marcus Rohrbach, Trevor Darrell, and Dan Klein (2016). Neural Module Networks. In

    Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

  • Arnold and Zuberbühler (2006a) Arnold, Kate and Klaus Zuberbühler (2006a). Language Evolution: Semantic Combinations in Primate Calls. Nature, 441: 303.
  • Arnold and Zuberbühler (2006b) Arnold, Kate and Klaus Zuberbühler (2006b). The Alarm-Calling System of Adult Male Putty-Nosed Monkeys, Cercopithecus nictitans martini. Animal Behaviour, 72: 643–653.
  • Arnold and Zuberbühler (2008) Arnold, Kate and Klaus Zuberbühler (2008). Meaningful Call Combinations in a Non-Human Primate. Current Biology, 18: R202–R203.
  • Arnold and Zuberbühler (2013) Arnold, Kate and Klaus Zuberbühler (2013). Female Putty-Nosed Monkeys Use Experimentally Altered Contextual Information to Disambiguate the Cause of Male Alarm Calls. PLOS ONE, 8(6): e65660.
  • Bahdanau et al. (2018) Bahdanau, Dzmitry, Shikhar Murty, Michael Noukhovitch, Thien Huu Nguyen, Harm de Vries, and Aaron Courville (2018). Systematic generalization: What is required and can it be learned? arXiv:1881.12889v1.
  • Barrett (2006) Barrett, Jeffrey (2006). Numerical Simulations of the Lewis Signaling Game: Learning Strategies, Pooling Equilibria, and Evolution of Grammar. Technical Report, Institute for Mathematical Behavioral Science.
  • Barrett (2007) Barrett, Jeffrey (2007). Dynamic Partitioning and the Conventionality of Kinds. Philosophy of Science, 74: 527–546.
  • Barrett (2009) Barrett, Jeffrey (2009). The Evolution of Coding in Signaling Games. Theory and Decision, 67: 223–237.
  • Barrett (2016) Barrett, Jeffrey (2016). On the Evolution of Truth. Erkenntnis, 81: 1323–1332.
  • Barrett (2017) Barrett, Jeffrey (2017). Truth and Probability in Evolutionary Games. Journal of Experimental and Theoretical Artificial Intelligence, 29(1): 219–225.
  • Barrett and LaCroix (2018) Barrett, Jeffrey A. and Travis LaCroix (2018). Epistemology and the structure of language. Unpublished Manuscript. August, 2018. PDF File.
  • Barrett et al. (2018) Barrett, Jeffrey A., Brian Skyrms, and Calvin Cochran (2018). Hierarchical Models for the Evolution of Compositional Language. Unpublished Manuscript. May, 2018. PDF File.
  • Barrett et al. (2017) Barrett, Jeffrey A., Brian Skyrms, and Aydin Mohseni (2017). Self-Assembling Networks. British Journal for the Philosophy of Science. Forthcoming.
  • Beck (2017) Beck, Jacob (2017). Can Bootstrapping Explain Concept Learning? Cognition, 158: 110–121.
  • Berwick and Chomsky (2011) Berwick, Robert and Noam Chomsky (2011). The Biolinguistic Program: The Current State of its Development. In Sciullo, A. M. Di and C. Boeckx, editors, The Biolinguistic Enterprise: New Perspectives on the Evolution and Nature of the Human Language Faculty, pages 19–41. Oxford University Press, Oxford.
  • Berwick (1998) Berwick, Robert C. (1998). Language Evolution and the Minimalist Program: The Origins of Syntax. In Hurford, James R., Michael Studdert-Kennedy, and Chris Knight, editors, Approaches to the Evolution of Language: Social and Cognitive Bases, pages 320–340. Cambridge University Press, Cambridge.
  • Berwick and Chomsky (2016) Berwick, Robert C. and Noam Chomsky (2016). Why Only Us: Language and Evolution. The MIT Press, Cambridge, MA.
  • Berwick et al. (2013a) Berwick, Robert C., Angela D. Friederici, Noam Chomsky, and Johan J. Bolhuis (2013a). Evolution, Brain, and the Nature of Language. Trends in Cognitive Sciences, 17(2): 89–98.
  • Berwick et al. (2013b) Berwick, Robert C., Marc D. Hauser, and Ian Tattersall (2013b). Neanderthal language? Just-So Stories Take Center Stage. Frontiers in Psychology, 4: 1–2.
  • Bickerton (1990) Bickerton, Derek (1990). Language and Species. University of Chicago Press, Chicago.
  • Bickerton (1998) Bickerton, Derek (1998). Catastrophic Evolution: The Case for a Single Step from Protolanguage to Full Human Language. In Hurford, James R., Michael Studdert-Kennedy, and Chris Knight, editors, Approaches to the Evolution of Language: Social and Cognitive Bases, pages 341–58. Cambridge University Press, Cambridge.
  • Bolhuis et al. (2014) Bolhuis, Johan J., Ian Tattersall, Noam Chomsky, and Robert C. Berwick (2014). How Could Language Have Evolved? PLOS Biology, 12(8): e1001934.
  • Bush and Mosteller (1955) Bush, Robert R. and Frederick Mosteller (1955). Stochastic Models for Learning. John Wiley & Sons, Oxford.
  • Carey (2004) Carey, Susan (2004). Bootstrapping and the Origin of Concepts. Daedalus, 133(1): 59–68.
  • Carey (2009a) Carey, Susan (2009a). The Origin of Concepts. Oxford Series in Cognitive Development. Oxford University Press, Oxford.
  • Carey (2009b) Carey, Susan (2009b). Where our Number Concepts Come From. The Journal of Philosophy, 106(4): 220–254.
  • Carey (2011a) Carey, Susan (2011a). Author’s Response: Concept Innateness, Concept Continuity, and Bootstrapping. Behavioral and Brain Sciences, 34(3): 152–162.
  • Carey (2011b) Carey, Susan (2011b). Precis of the Origin of Concepts. Behavioral and Brain Sciences, 34(3): 113–124.
  • Carey (2014) Carey, Susan (2014). On Learning New Primitives in the Language of Thought: Reply to Rey. Mind & Language, 29(2): 133–166.
  • Carstairs-McCarthy (1999) Carstairs-McCarthy, Andrew (1999). The Origins of Complex Language. Oxford University Press, Oxford.
  • Choi et al. (2018) Choi, Edward, Angeliki Lazaridou, and Nando de Freitas (2018). Compositional obverter communication learning from raw visual input. arXiv preprint arXiv:1804.02341.
  • Chomsky (2002) Chomsky, Noam (2002). On Nature and Language. Cambridge University Press, Cambridge.
  • Chomsky (2005) Chomsky, Noam (2005). Three Factors in Language Design. Linguistic Inquiry, 36: 1–22.
  • Chomsky (2010) Chomsky, Noam (2010). Some Simple Evo Devo Theses: How True Might They Be for Langauge? In Larson, Richard K., Viviane Déprez, and Hiroko Yamakido, editors, The Evolution of Human Language: Biolinguistic Perspectives, pages 148–162. Cambridge University Press, Cambridge.
  • Christiansen and Kirby (2003) Christiansen, Morten H. and Simon Kirby (2003). Language Evolution: The Hardest Problem in Science? In Christiansen, Morten H. and Simon Kirby, editors, Language Evolution, pages 1–15. Oxford University Press, Oxford.
  • Culicover and Jackendoff (2005) Culicover, Peter W. and Ray Jackendoff (2005). Simpler Syntax. Oxford University Press, New York.
  • Das et al. (2017) Das, Abhishek, Satwik Kottur, José MF Moura, Stefan Lee, and Dhruv Batra (2017). Learning cooperative visual dialog agents with deep reinforcement learning. In Proceedings of the IEEE International Conference on Computer Vision, pages 2951–2960.
  • Di Sciullo (2011) Di Sciullo, Anna Maria (2011). A Biolinguistic Approach to Morphological Variation. In Sciullo, A. M. Di and C. Boeckx, editors, The Biolinguistic Enterprise: New Perspectives on the Evolution and Nature of the Human Language Faculty, pages 305–326. Oxford University Press, Oxford.
  • Di Sciullo (2013) Di Sciullo, Anna Maria (2013). Exocentric Compounds, Language and Proto-Language. Language and Information Society, 20: 1–26.
  • Erev and Roth (1998) Erev, Ido and Alvin E. Roth (1998). Predicting How People Play Games: Reinforcement Learning in Experimental Games with Unique, Mixed Strategy Equilibria. The American Economic Review, 88(4): 848–881.
  • Evtimova et al. (2017) Evtimova, Katrina, Andrew Drozdov, Douwe Kiela, and Kyunghyun Cho (2017). Emergent communication in a multi-modal, multi-step referential game. arXiv preprint arXiv:1705.10369.
  • Fitch (2004) Fitch, W. Tecumseh (2004). Kin Selection and ‘Mother Tongues’: A Neglected Component in Language Evolution. In Oller, D. K. and U. Griebel, editors, Evolution of Communication Systems: A Comparative Approach, pages 275–296. MIT Press, Cambridge, MA.
  • Fitch (2010) Fitch, W. Tecumseh (2010). The Evolution of Language. Cambridge University Press, Cambridge.
  • Fodor and Pylyshyn (1988) Fodor, Jerry A. and Zenon W. Pylyshyn (1988). Connectionism and Cognitive Architecture: A Critical Analysis. Cognition, 28(1): 3–71.
  • Franke (2014) Franke, Michael (2014). Creative Compositionality from Reinforcement Learning in Signaling Games. In Cartmill, Erica A., Seén Roberts, Heidi Lyn, and Hannah Cornish, editors, The Evolution of Language: Proceedings of the 10th International Conference, volume 10 of Evolang, pages 82–89. World Scientific, Singapore.
  • Franke (2016) Franke, Michael (2016). The Evolution of Compositionality in Signaling Games. Journal of Logic, Language and Information, 25(3): 355–377.
  • Gaunt et al. (2016) Gaunt, Alexander L., Marc Brockschmidt, Nate Kushman, and Daniel Tarlow (2016). Differentiable Programs with Neural Libraries. In Proceedings of the 34th International Conference on Machine Learning. arXiv:1611.02109.
  • Givón (1979) Givón, Talmy (1979). On Understanding Grammar. Academic Press, New York.
  • Givón (2002a) Givón, Talmy (2002a). Bio-Linguistics: The Santa Barbara Lectures. John Benjamins Publishing Company, Amsterdam.
  • Givón (2002b) Givón, Talmy (2002b). The Visual Information-Processing System as an Evolutionary Precursor to Human Language. In Givón, Talmy and Bertram F. Malle, editors, The Evolution of Language Out of Pre-Language, volume 53 of Typological Studies in Language, pages 3–50. John Benjamins, Amsterdam.
  • Givón (2009) Givón, Talmy (2009). The Genesis of Syntactic Complexity: Diachrony, Ontogeny, Neuro-cognition, Evolution. John Benjamins Publishing Company, Amsterdam/Philadelphia.
  • Hauser et al. (2002) Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch (2002). The Faculty of Language: What Is It, Who Has It, and How Did It Evolve? Science, 298: 1569–1579.
  • Havrylov and Titov (2017) Havrylov, Serhii and Ivan Titov (2017). Emergence of language with multi-agent games: Learning to communicate with sequences of symbols. In Advances in neural information processing systems, pages 2149–2159.
  • Heine and Kuteva (2007) Heine, Bernd and Tania Kuteva (2007). The Genesis of Grammar: A Reconstruction. Oxford University Press, Oxford.
  • Hornstein (2008) Hornstein, Norbert (2008). A Theory of Syntax: Minimal Operations and Universal Grammar. Cambridge University Press, Cambridge.
  • Hu et al. (2017) Hu, Ronghang, Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Kate Saenko (2017). End-to-End Module Networks for Visual Question Answering. In Proceedings of 2017 IEEE International Conference on Computer Vision. arXiv:1704.05526.
  • Hurford (2007) Hurford, James R. (2007). Language in the Light of Evolution I: The Origins of Meaning. Oxford University Press, Oxford.
  • Hurford (2012) Hurford, James R. (2012). Language in the Light of Evolution II: The Origins of Grammar. Oxford University Press, Oxford.
  • Huttegger (2007) Huttegger, Simon M. (2007). Evolutionary Explanations of Indicatives and Imperatives. Erkenntnis, 66: 409–436.
  • Jackendoff (1999) Jackendoff, Ray S. (1999). Possible Stages in the Evolution of the Language Capacity. Trends in Cognitive Sciences, 3: 272–279.
  • Jackendoff (2002) Jackendoff, Ray S. (2002). Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford University Press, Oxford.
  • Johnson et al. (2016) Johnson, Justin, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C. Lawrence Zitnick, and Ross Girshick (2016). Clevr: A diagnostic dataset for compositional language and elementary visual reasoning. In Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). arXiv:1612.06890.
  • Johnson et al. (2017) Johnson, Justin, Bharath Hariharan, Laurens van der Maaten, Judy Hoffman, Li Fei-Fei, C. Lawrence Zitnick, and Ross Girshick (2017). Inferring and executing programs for visual reasoning. In Proceedings of 2017 IEEE International Conference on Computer Vision.
  • Kamp and Partee (1995) Kamp, Hans and Barbara Partee (1995). Prototype Theory and Compositionality. Cognition, 57: 129–191.
  • Kottur et al. (2017) Kottur, Satwik, José M. F. Moura, Stefan Lee, and Dhruv Batra (2017). Natural language does not emerge ’naturally’ in multi-agent dialog. CoRR, abs/1706.08502.
  • LaCroix (2019a) LaCroix, Travis (2019a). Accounting for polysemy and role asymmetry in the evolution of compositional signals. Unpublished Manuscript. May, 2019. PDF File.
  • LaCroix (2019b) LaCroix, Travis (2019b). Evolutionary explanations of simple communication: Signalling games and their models. Journal for General Philosophy of Science / Zeitschrift für allgemeine Wissenschaftstheorie. Forthcoming.
  • LaCroix (2019c) LaCroix, Travis (2019c). If gradualism is the correct approach ot language origins, then compositionality is not a plausible explanatory target. Unpublished Manuscript. September, 2019. PDF File.
  • LaCroix (2019d) LaCroix, Travis (2019d). Saltationist versus gradualist approaches to language origins: A critical discussion. Unpublished Manuscript. March, 2019. PDF File.
  • LaCroix (2019e) LaCroix, Travis (2019e). Using logic to evolve more logic: Composing logical operators via self-assembly. British Journal for the Philosophy of Science. Forthcoming.
  • LaCroix (2020) LaCroix, Travis (2020). Complex Signals: Reflexivity, Hierarchical Structure, and Modular Composition. PhD dissertation, University of California, Irvine.
  • Lake and Baroni (2017) Lake, Brenden M. and Marco Baroni (2017). Generalization Without Systematicity: On the Compositional Skills of Sequence-to-Sequence Recurrent Networks. arXiv:1711.00350.
  • Lazaridou et al. (2018) Lazaridou, Angeliki, Karl Moritz Hermann, Karl Tuyls, and Stephen Clark (2018). Emergence of linguistic communication from referential games with symbolic and pixel input. CoRR, abs/1804.03984.
  • Lazaridou et al. (2016) Lazaridou, Angeliki, Alexander Peysakhovich, and Marco Baroni (2016). Multi-agent cooperation and the emergence of (natural) language. arXiv preprint arXiv:1612.07182.
  • Lewis (1969) Lewis, David (2002/1969). Convention: A Philosophical Study. Blackwell, Oxford.
  • Lightfoot (1991) Lightfoot, David (1991). Subjacency and Sex. Language & Communication, 11(1–2): 67–69.
  • Margolis and Laurence (2008) Margolis, Eric and Stephen Laurence (2008). How to Learn the Natural Numbers: Inductive Inference and the Acquisition of Number Concepts. Cognition, 106(2): 924–939.
  • Margolis and Laurence (2011) Margolis, Eric and Stephen Laurence (2011). Beyond the Building Blocks Model. Behavioral and Brain Sciences, 34(3): 139–140.
  • Martínez (2015) Martínez, Manolo (2015). Deception in Sender-Receiver Games. Erkenntnis, 80(1): 215–227.
  • Miyagawa (2017) Miyagawa, Shigeru (2017). Integration Hypothesis: A Parallel Model of Language Development in Evolution. In Watanabe, S., M. Hofman, and T. Shimizu, editors, Evolution of the Brain, Cognition, and Emotion in Vertebrates, Brain Science, pages 225–247. Springer, Tokyo.
  • Miyagawa et al. (2015) Miyagawa, Shigeru, Shiro Ojima, Robert C. Berwick, and Kazuo Okanoya (2015). The Integration Hypothesis of Human Language Evolution and the Nature of Contemporary Languages. Frontiers in psychology, 39: 564.
  • Moro (2008) Moro, Andrea (2008). The Boundaries of Babel: The Brain and the Enigma of Impossible Languages. The MIT Press, Cambridge, MA.
  • Neftci and Averbeck (2019) Neftci, Emre O. and Bruno B. Averbeck (2019). Reinforcement Learning in Artificial and Biological Systems. Nature Machine Intelligence, 1: 133–143.
  • Newmeyer (1991) Newmeyer, Frederick J. (1991). Functional Explanation in Linguistics and the Origin of Language. Language and Communication, 11: 1–28.
  • Newmeyer (1998) Newmeyer, Frederick J. (1998). On the supposed ‘counterfunctionality’ of Universal Grammar: Some evolutionary implications. In Hurford, James R., Michael Studdert-Kennedy, and Chris Knight, editors, Approaches to the Evolution of Language: Social and Cognitive Bases, pages 305–319. Cambridge University Press, Cambridge.
  • Newmeyer (2005) Newmeyer, Frederick J. (2005). Possible and Probable Languages: A Generative Perspective on Linguistic Typology. Oxford University Press, Oxford.
  • Nowak and Krakauer (1999) Nowak, Martin A. and David C. Krakauer (1999). The Evolution of Language. Proceedings of the National Academy of Sciences, 96: 8028–8033.
  • O’Connor (2014) O’Connor, Cailin (2014). The Evolution of Vagueness. Erkenntnis, 79(4): 707–727.
  • Pemantle and Skyrms (2004) Pemantle, Robin and Brian Skyrms (2004). Network Formation by Reinforcement Learning: the Long and the Medium Run. Mathematical Social Sciences, 48: 315–327.
  • Piantadosi et al. (2012) Piantadosi, Steven T., Joshua B. Tenenbaum, and Noah D. Goodman (2012). Bootstrapping in a Language of Thought: A Formal Model of Numerical Concept Learning. Cognition, 123(2): 199–217.
  • Piattelli-Palmarini (2010) Piattelli-Palmarini, Massimo (2010). What Is Language, That It May Have Evolved, and What Is Evolution, That It May Apply to Language? In Larson, Richard K., Viviane Déprez, and Hiroko Yamakido, editors, The Evolution of Human Language: Biolinguistic Perspectives, pages 148–162. Cambridge University Press, Cambridge.
  • Piattelli-Palmarini and Uriagereka (2004) Piattelli-Palmarini, Massimo and Juan Uriagereka (2004). The Immune Syntax: The Evolution of the Language Virus. In Jenkins, Lyle, editor, Variation and Universals in Biolinguistics, volume 62 of Linguistic Variations, chapter 14, pages 341–377. Elsevier, Oxford.
  • Piattelli-Palmarini and Uriagereka (2011) Piattelli-Palmarini, Massimo and Juan Uriagereka (2011). A Geneticist’s Dream, a Linguist’s Nightmare: The Case of FOXP2. In Sciullo, Anna Maria Di and Cedric Boeckx, editors, The Biolinguistic Enterprise: New Perspectives on the Evolution and Nature of the Human Language Faculty, Oxford Studies in Biolinguistics, chapter 5, pages 100–125. Oxford University Press, Oxford.
  • Pinker and Bloom (1990) Pinker, Steven and Paul Bloom (1990). Natural Language and Natural Selection. Behavioral and Brain Sciences, 13: 707–726.
  • Progovac (2006) Progovac, Ljiljana (2006). The Syntax of Nonsententials: Small Clauses and Phrases at the Root. In Progovac, Ljiljana, Kate Paesani, Eugenia Casielles-Suarez, and Ellen Barton, editors, The Syntax of Nonsententials: Multidisciplinary Perspectives, volume 93 of Linguistik Aktuell/Linguistics Today, pages 33–71. John Benjamins, Amsterdam.
  • Progovac (2009a) Progovac, Ljiljana (2009a). Layering of Grammar: Vestiges of Protosyntax in Present-Day Languages. In Sampsom, Geoffrey, David Gil, and Peter Trudgill, editors, Language Complexity as an Evolving Variable, Studies in the Evolution of Language. Oxford University Press, New York.
  • Progovac (2009b) Progovac, Ljiljana (2009b). Sex and Syntax: Subjacency Revisited. Biolinguistics, 3(2–3): 305–336.
  • Progovac (2013) Progovac, Ljiljana (2013). Rigid Syntax, Rigid Sense: Absolutives/Unaccusatives as Evolutionary Precursors. In Franks, Steven, Markus Dickinson, George Fowler, Melissa Witcombe, and Ksenia Zanon, editors, Proceedings of Formal Approaches to Slavic Linguistics (FASL), The Third Indiana Meeting, Bloomington, IN., pages 246–259. Michigan Slavic Publications, Ann Arbor.
  • Progovac (2015) Progovac, Ljiljana (2015). Evolutionary Syntax. Oxford University Press, Oxford.
  • Progovac (2019) Progovac, Ljiljana (2019). A Critical Introduction to Language Evolution: Current Controversies and Future Prospects. Springer, Berlin.
  • Rescorla and Wagner (1972) Rescorla, Robert A. and Allan R. Wagner (1972). A Theory of Pavlovian Conditioning: Variations in the Effectiveness of Reinforcement and Nonreinforcement. In Black, A. H. and W. F. Prokasy, editors, Classical Conditioning II: Current Research and Theory, pages 64–99. Appleton-Century-Crofts, New York.
  • Roth and Erev (1995) Roth, Alvin and Ido Erev (1995). Learning in Extensive Form Games: Experimental Data and Simple Dynamical Models in the Intermediate Term. Games and Economic Behavior, 8: 164–212.
  • Shea (2009) Shea, Nicholas (2009). New Concepts Can Be Learned. Biology and Philosophy, 26(1): 129–139.
  • Skyrms (2010) Skyrms, Brian (2010). Signals: Evolution, Learning, & Information. Oxford University Press, Oxford.
  • Skyrms (1996) Skyrms, Brian (2014/1996). Evolution of the Social Contract. Cambridge University Press, Cambridge.
  • Skyrms and Barrett (2018) Skyrms, Brian and Jeffrey A. Barrett (2018). Propositional Content in Signals. Studies in the History and Philosophy of Science C. Forthcoming.
  • Steinert-Threlkeld (2014) Steinert-Threlkeld, Shane (2014). Learning to use function words in signaling games. In Lorini, Emiliano and Laurent Perrussel, editors, Proceedings of Information Dynamics in Artificial Societies, (IDAS-14).
  • Steinert-Threlkeld (2016) Steinert-Threlkeld, Shane (2016). Compositional signaling in a complex world. Journal of Logic, Language, and Information, 25(3–4): 379–397.
  • Steinert-Threlkeld (2018) Steinert-Threlkeld, Shane (2018). Function Words and Context Variability. PSA 2018: The 26th Biennial Meeting of the Philosophy of Science Association, 1–4 November 2018.
  • Szabó (2012) Szabó, Zoltán Gendler (2012). The Case for Compositionality. In Hinzen, Wolfram, Edouard Machery, and Markus Werning, editors, The Oxford Handbook of Compositionality, pages 64–80. Oxford University Press, Oxford.
  • Számadó and Szathmary (2006) Számadó, S. and E. Szathmary (2006). Selective Scenarios for the Emergence of Natural Language. Trends in Ecology & Evolution, 21(10): 555–561.
  • Tallerman (2007) Tallerman, Maggie (2007). Did our ancestors speak a holistic protolanguage? Lingua, 117(3): 579–604.
  • Tallerman (2013a) Tallerman, Maggie (2013a). Join the Dots: A Musical Interlude in the Evolution of Language? Journal of Linguistics, 49: 455–487.
  • Tallerman (2013b) Tallerman, Maggie (2013b). Kin Selection, Pedagogy, and Linguistic Complexity: Whence Protolanguage? In Botha, Rudolf and Martin Everaert, editors, The Evolutionary Emergence of Language, page 77–96. Oxford University Press, Oxford.
  • Tallerman (2014a) Tallerman, Maggie (2014a). Is the Syntax Rubicon More of a Mirage? A Defence of Pre-Syntactic Protolanguage. In Cartmill, Erica A., Seén Roberts, Heidi Lyn, and Hannah Cornish, editors, The Evolution of Language: Proceedings of the 10th International Conference, volume 10 of Evolang, pages 318–325. World Scientific, Vienna.
  • Tallerman (2014b) Tallerman, Maggie (2014b). The Evolutionary Origins of Syntax. In Carnie, Andrew, Yosuke Sato, and Daniel Siddiqi, editors, The Routledge Handbook of Syntax, pages 446–462. Routledge, London.
  • Tallerman and Gibson (2011) Tallerman, M. and K. Gibson (2011). The Oxford Handbook of Language Evolution. Oxford University Press, Oxford.
  • Wagner (2014) Wagner, Elliott O. (2014). Conventional Semantic Meaning in Signalling Games with Conflicting Interests. British Journal for the Philosophy of Science, 66(4): 751–773.
  • Wagner et al. (2003) Wagner, Kyle, James A. Reggia, Juan Uriagereka, and Gerald S. Wilkinson (2003). Progress in the Simulation of Emergent Communication and Language. Adaptive Behavior, 11(1): 37–69.
  • Yang (2013) Yang, C. (2013). Ontogeny and phylogeny of language. Proceedings of the National Academy of Sciences of the United States of America, 110(16): 6323–6327.
  • Zollman (2011) Zollman, Kevin J. S. (2011). Separating Directives and Assertions Using Simple Signaling Games. The Journal of Philosophy, 108(3): 158–169.
  • Zollman et al. (2012) Zollman, Kevin J. S., Carl T. Bergstrom, and Simon M. Huttegger (2012). Between Cheap and Costly Signals: The Evolution of Partially Honest Communication. Proceedings of the Royal Society B: Biological Sciences, 280(1750): 20121878.
  • Zuberbühler (2001) Zuberbühler, Klaus (2001). Predator-Specific Alarm Calls in Campbell’s Monkeys, Cercopithecus Campbelli. Behavioural Ecology and Sociobiology, 50: 414–422.
  • Zuberbühler (2002) Zuberbühler, Klaus (2002). A Syntactic Rule in Forest Monkey Communication. Animal Behaviour, 63(2): 293–299.